EBITDA from −1.8% to 4.6% in nine months: how to make a restaurant profitable when record sales hide the leak, using the Restaurant Model Canvas

Why did a grill house growing 19% still close the year in the red?

Because the channel MIX was wrong, not the demand. It billed 738 thousand dollars over twelve months, climbed 19% against the prior year, and still closed with EBITDA of −1.8%:

every extra dollar came through a channel taking 27% commission on a ticket whose contribution margin was 34%. Seven points were left to cover rent, payroll and utilities. The outside benchmark makes noise against that: Statista puts restaurant net margin between 3% and 9%, and full-service between 3% and 5%, while listed chains run 12%-13% operating margin after taxes according to the Restaurant Financial Benchmarks 2026 from WhippleWood CPAs. This grill house was not living in the low tail of that distribution. It was living outside it, lending working capital to three apps. Ranking the leaks by size, rather than by accounting habit, changed the whole treatment. The gap between real food cost (36.2%) and theoretical (30.3%) reached 5.9 points on an annual raw-material purchase of 267 thousand dollars: roughly 43 thousand dollars a year evaporating into waste, free pours and deliveries received without control.

Food cost was bad, and even so it was not the main leak

Serious figure. But aggregator commission consumed 121 thousand dollars over the same period, nearly three times more, and had never appeared as a separate line on any sheet of the business: it came netted against the platform's weekly deposit, invisible. A trade warning belongs here, because payroll also squeezes: Toast reported in 2024 that labor already exceeds 25% of a restaurant's expenses, up from 23% in 2021. Three fronts, one possible order. Splitting the P&L by channel —dining room, pickup, own delivery and each aggregator— was the first move of the MASTERESTAURANT method, and we did it with the Channel Margin Map, the piece of the costing kit that Diego F. Parra applies in the firm's audits. The mechanics are cold: take the real average ticket of each channel (21.40 dollars in the dining room against 16.80 in delivery), subtract the theoretical plate cost, the commission, the packaging and the kitchen time it occupies, and contribution shows up in dollars per order, not in percentage.

The tool we used: the Masterestaurant Channel Margin Map

That is where the business saw itself straight on: 2.7 dollars of contribution per aggregator order against 7.3 dollars per cover in the dining room. Nobody argues with that table. The order of treatment matters more than any single lever, and I will take a firm position: raising prices first would have been the expensive mistake. Follow it through to its consequence. With 61% of orders arriving through platforms, an 8% menu increase hands 27 cents of every new dollar to the aggregator, exposes the customer to price comparison inside the same app —where the competitor next door stays cheap— and leaves the dining-room ticket more expensive for no visible reason. We would recover maybe 14 thousand dollars gross and lose volume in the channel that was paying rent. So we moved the mix first, then closed the theoretical gap with standardized recipes and receiving control, and only in month six touched eight menu prices.

Mix first, theoretical cost second, and the menu only at the end

Three of them downward. An owned channel was built before touching the aggregator, never the other way around: switching platforms off without an alternative destination is cash suicide. Months one and two, own online ordering with a short menu of 22 profitable dishes and delivery by two contracted riders at 4.10 dollars per drop against the 5.60 average the same delivery cost through the app. Month three, QR inserts in the 1,900 monthly packages already going out via aggregator, offering a dessert on the first direct order. Month four, pickup with a guaranteed 12-minute window and a 6% discount. Month five, we shut off DiDi Food, the platform with the worst contribution. Statista holds that 3%-9% sector net margin as a realistic yardstick; the internal target was 7% in fourteen months, not a heroic number. Margin changed, revenue did not, and that is the useful reading of the whole case.

The results at fourteen months, with the case figures

Annualized sales landed at 761 thousand dollars, barely 3.1% up, essentially flat against the previous year's 19%. Now look at the rest: aggregator dependency from 61% to 28% of orders, commission paid from 121 thousand to 47 thousand dollars, real food cost from 36.2% to 31.4%, EBITDA from −1.8% to +6.9%, and operating cash covering 71 days of payroll with no credit line. Contribution per own-delivery order settled at 5.9 dollars. Selling the same left 66 thousand dollars more in operating profit. And there is an asset effect almost nobody computes: Sofer Advisors places the sale of an independent single-location restaurant between 1.5x and 3x SDE, so that repaired EBITDA also revalued the business. Each band has a different first step this week, and none of them costs money. Under 500 thousand dollars a year: sit down with the last 90 days of platform deposits and write commission as its own line on your sheet; a number the owner had never seen almost always shows up.

Transferable lessons by annual revenue band

From 500 thousand to 1 million, this grill house's case: compute contribution in DOLLARS per order for each channel before touching prices. From 1 to 5 million: audit the packaging and the kitchen time the delivery channel consumes, because a saturated kitchen working for free usually hides right there. Above 5 million: renegotiate commission by volume with a written exit date, and build your own delivery brand. Above 10 million, group or chain —the archetype of the media chef with six themed locations and a product line—: consolidate the P&L by channel at group level before doing it by store, because the leak offsets across sites and vanishes from the report. I would not expect this result in three contexts, and I prefer to say so before someone copies the recipe. One: ghost kitchens and operations born 100% on platform, with no dining room and no geographic clientele of their own, where the aggregator is not an expensive channel but the entire business —a market Credence Research valued at 72,060 million dollars in 2024, and one that plays by different physics.

Limits of this case

Two: markets where a single app concentrates delivery demand and the customer has no habit of ordering direct; switching off there does not free margin, it erases sales. Three: businesses with contribution margin below 25%, typical of high input-cost menus, because the arithmetic simply leaves no remainder after commission, whatever the mix. And a warning about survivorship bias: this owner endured five months of transition with tight cash. Whoever cannot finance that bridge needs a different plan. Food cost was not the main leak, even though food cost was bad. At 36.2% actual against 30.3% theoretical, that 5.9-point gap burned roughly 43 thousand dollars a year on 267 thousand of annual raw material purchases. Serious, certainly. Yet aggregator commission took 121 thousand dollars over the same window, nearly three times as much, and nobody had ever written it on a sheet as its own cost line.

Where the leak actually was?

A restaurant can run impeccable food cost and still lose money if it gives away margin at the order's front door. The reverse happens too.

That is why treatment sequencing went channel mix first, theoretical cost second, prices only at the end: had we opened by raising menu prices, the aggregator would have kept 27% of the increase while the customer compared us against competitors on the same app grid. Labor cost at 32.2% looked bloated until we crossed it with 94% annual turnover. Training a grill cook costs six to eight weeks of reduced productivity, and this kitchen had spent two years training people who quit before they paid back. Toast reported in 2024 that payroll now exceeds 25% of restaurant expenses, up from 23% in 2021, as covered by Restaurant Dive; the issue here was never the hourly rate, it was paying three times for the same position within one year.

Where the leak actually was — in practice?

The deferred P&L hid the cash flow. An income statement arriving on the 22nd is not accounting, it is archaeology.

By the time the owner saw that March closed in the red, April's purchasing and scheduling had already been set with the same logic that sank March, and that three-week lag repeated twelve times a year. And this trade never fully resolves one tension: the aggregator is expensive, and it also brings customers the neighborhood would not have delivered. Switching it off was never the answer. We stopped growing in it instead, freezing platform volume while the direct channel climbed, until relative weight dropped from 61% to 38% without losing a single order in absolute terms.